Admixture cv error

Question

[s.bourne]

Hi all,

When doing Admixture and outputting cv error rate for a range of k values, one of the k values keeps on outputting "nan" as the error rate. It works for the other k values, but not for this one. Has this happened to anyone else?

Cheers,
Steve

[jcolella.jc]

Hi Steve, did you ever figure this out?? I get the same issue every time I run K=3 on my data.

Open to suggestions!

[jenn.drummond]

Same here. Here's some sample output from the bottom of the file. I'm not so worried about the numbers as I am about the fact that the job isn't finishing. It's been running for four hours and all the other versions are done, including ones with higher K values.

586 (QN/Block)  Elapsed: 70.414 Loglikelihood: -nan     (delta): -nan
587 (QN/Block)  Elapsed: 70.526 Loglikelihood: -nan     (delta): -nan
588 (QN/Block)  Elapsed: 70.537 Loglikelihood: -nan     (delta): -nan
589 (QN/Block)  Elapsed: 70.47  Loglikelihood: -nan     (delta): -nan

[blair.chr]

I'm running into the same issue. I am receiving 'nan' for Loglikelihood and delta for K=6 only. K=1-5 works fine as does K=7. This is quite odd. Any thoughts?

Chris

Answer 1

[chengcw]

You may want to run with admixture32 instead. Hope it helps.

Answer 2

[Declan]

I had this issue today and was able to resolve it by changing the random seed Admixture uses by adding -s time (which sets the random seed based on the current time) to my command so it looks something like this:

admixture --cv -s time inputFile.ped 6

I'm not sure why this works, but so far it has fixed the issue for me. I previously had tried restarting the K=6 run several times and the seed was always the same, so it always gave me the nan Loglikelihood and delta values. I figured trying a different starting point might fix things.

-Declan

Edit: While continuing to run Admixture with other data, the issue still sometimes persists with this solution in place. However, stopping the run and restarting to get a new seed seems to help. It's a bit hit or miss.